Molecular Probe-Guided Surface-Enhanced Raman Scattering Sensing Interfaces and Multiview Feature Fusion for Interpretable Serum Fingerprinting in Prostate Disease Classification.

Abstract

Accurate discrimination between prostate cancer (PCa) and benign prostatic hyperplasia (BPH) remains clinically challenging. Here, we developed a multiprobe serum surface-enhanced Raman scattering (SERS) sensing platform comprising Ag NPs (Ag), 4-mercaptobenzoic acid (MBA)@Ag, 4-mercaptophenylboronic acid (MPBA)@Ag, and 4-aminothiophenol (ATP)@Ag, together with a molecular probe-guided spectral feature selection and multiview fusion (MPGSF) strategy for classifying PCa, BPH, and healthy group samples. The probe-modified substrates provided distinct surface-chemical interfaces and reporter-peak modulation patterns. MPGSF selected Raman-shift windows from probe-induced differential responses by integrating the peak intensity, local signal-to-noise ratio, repeated-measurement stability, and extracted sample-level multiview features from measured spectra. In sample-level cross-validation, the MPGSF-random forest (RF) model achieved 92.44% accuracy and 92.13% macro-F1, outperforming single-view and full-spectrum concatenation baselines. In the independent batch hold-out validation set, it maintained 90.00% accuracy and 89.95% macro-F1. Among PCa samples with prostate-specific membrane antigen positron emission tomography molecular-imaging reference information, the PCa recognition consistency was 39/43 (90.7%). Feature-contribution analysis showed that all four SERS views contributed to classification, with high-contribution windows located in serum-response regions with potential biochemical relevance. MPGSF-guided multiprobe SERS provides an interpretable analytical framework for minimally invasive auxiliary classification of prostate diseases.